This chapter documents the very local human capital spillovers within cities. College employ- ment density raises the productivity of workers around them, while non-college employment density is associated with congestive forces that decrease the productivity of workers around them. Moreover, both of these effects only occur within 2-3 miles, so the reach is limited. I started with some stylized facts on the geographical distribution of workplaces within cities that the literature has not yet shown. Then, I presented both descriptive evidence of this phenomenon and framed the results using an economic geography model that takes into account commuting, agglomeration, and human capital spillovers across tracts. These rapidly decaying human capital spillovers are consistent with previous papers estimating local spillovers of density of all workers (Ahlfeldt et al. 2015), of certain industries (Arzhagi and Henderson 2006, Berkes and Gaetani 2019), and of human capital (Fu 2007, Rosenthal and Strange 2008).
enhancing density and re-zoning to allow for higher-quality office spaces to attract high-skill jobs into certain neighborhoods. By varying the level of density of development and increase in college share, I find that if development follows the desired density alone, there will be a fall in average productivity due to an increase in non-college workplaces in the area. When coupled with policies to encourage the creation of more high-skill jobs, these policies create spillover effects that benefit beyond the targeted locations.
The estimates of local spillover effects of human capital presented in this chapter are relevant to urban policymakers who must consider where to place employment centers and the effect density on the surrounding areas. The parameter estimates and model framework presented in this chapter can inform policymakers as to what would be the general equi- librium effects of human capital density, and how they affect the targeted and surrounding areas. However, the fast decay of these effects will also limit the reach of density and growth in the city. Especially worrisome is the recent trend in the loss of accessibility to job clusters for the urban poor as gentrification picks up across the United States (Kneebone and Holmes 2015). These fast-decaying spillovers imply that the loss of accessibility to jobs near high human capital centers will exacerbate the welfare loss of the urban poor in the coming years. I end this chapter with some avenues for extension. First, better individual-level data with finer geography and a panel structure, such as the underlying LEHD data, could be used to better empirically estimate the human capital spillover effects. Second, incorporating the endogenous response of developers and firms in a model with spatial spillovers will be important; unlike the counterfactual exercises presented in this chapter, such a model will better predict outcomes from changes in zoning. Third, exploring potential mechanisms behind human capital spillovers I document will provide the right context to interpret the estimates of this chapter.
1.A
Appendix Tables and Figures
Table 1.A.1: Robustness Checks for Different Sample Definitions
(1) (2) (3) (4)
Sample One Worker Two Workers Two Workers Full-time in HH One Full-time in HH
Log Col Own 0.102*** 0.104*** 0.0958*** 0.106*** (0.0197) (0.0200) (0.00702) (0.0105) Log Col 0-3 mi 0.0972*** 0.0610 0.0782*** 0.0805*** (0.0341) (0.0365) (0.0189) (0.0240) Log Col 3-5 mi -0.00863 0.0634** 0.0216 0.0602*** (0.0381) (0.0246) (0.0151) (0.0213) Log Col 5-10 mi -0.0320 -0.0371 -0.0186 -0.0535 (0.0372) (0.0289) (0.0200) (0.0381) Log Col 10-25 mi -0.0276 -0.00600 -0.00258 0.0500 (0.0647) (0.0397) (0.0395) (0.0395) Log Non Own -0.0735*** -0.0825*** -0.0778*** -0.0890***
(0.0192) (0.0205) (0.00802) (0.0116) Log Non 0-3 mi -0.107*** -0.0596 -0.0722*** -0.0712** (0.0375) (0.0414) (0.0212) (0.0278) Log Non 3-5 mi 0.00340 -0.0799*** -0.0295 -0.0740*** (0.0412) (0.0271) (0.0198) (0.0228) Log Non 5-10 mi 0.0608 0.0687* 0.0404* 0.0749* (0.0400) (0.0356) (0.0221) (0.0398) Log Non 10-25 mi 0.0405 -0.000457 0.0180 -0.0407 (0.0777) (0.0483) (0.0450) (0.0428) Constant 10.56*** 10.05*** 10.08*** 10.34*** (0.453) (0.579) (0.328) (0.426) Observations 9,848 10,617 24,591 21,565 R-squared 0.492 0.491 0.483 0.461
OLS specification with clustered standard errors at the employment tract level in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Column (1) uses only households with one worker in the sample. Column (2) uses only “primary earners” defined by the only full-time worker in households with two workers. Column (3) uses only households with 2 workers or less. Column (4) uses only full-time workers.
Table 1.A.2: Correlation of LODES Between Different Data Sources Employment Residence
vs CHTS vs ACS College 0.863 0.847 Non-college 0.713 0.791
Correlation coefficients of LODES dataset against different data sets. First column checks employment location against CHTS and second column checks residence location vs American Community Survey 2010. Both data restricts samples to workers (employeed).
Figure 1.A.1: OLS with Finer Spatial Definition
(a) OLS with Same Specification As Paper
(b) OLS Disaggregating 0-5 miles
Coefficient estimates and 95% confidence intervals from pooled OLS regression specification as in Table1.2
column (1) with varying distance ring definitions. Figure1.A.1aplots the specification from column (1) and Figure1.A.1bplots the specification with 0-3 miles and 4-5 miles disaggregated to 0-1, 1-2, 2-3, 3-4, 4-5 mile rings.
Figure 1.A.2: Over-Identification Check for Floor Space by Area
(a) Chicago (b) New York
(c) Los Angeles (d) Philadelphia
Over-identification checks for density of development separately by area. Chicago, New York, Philadelphia data are for the city boundaries. Los Angeles data is for the County of Los Angeles. Correlation coefficients for Chicago, New York, Los Angeles, Philadelphia are 0.48, 0.7, 0.7, 0.79, respectively.
CHAPTER 2
Do Youth Employment Programs Work? Evidence
from the New Deal
12.1
Introduction
Unemployment rates are typically highest among the young, particularly those from poor backgrounds and during recessions. At the height of the Great Recession, unemployment rates for those over age 25 peaked at 8.4% in 2010 but were as high as 19.6% for those aged 16-24 (US Bureau of Labor Statistics 2018). To address youth unemployment, government- run employment training programs specifically target young adults. However, the short run effects of these programs have been shown to be modest, at best, and there is very limited evidence of their effectiveness over the long run. There is also very limited evidence on the effects of these programs on non-labor market outcomes and on the mechanisms by which
1Anna Aizer, Brown University ([email protected]), Shari Eli, University of Toronto
([email protected]), Keyoung Lee, UCLA ([email protected]), and Adriana Lleras-Muney, UCLA ([email protected]). We are very grateful to many research assistants that worked on this project, especially to Ryan Boone, Taehoon Kang and Kyle Sherman. We have benefitted from comments from participants in the various conferences. We are particularly indebted to Rodrigo Pinto for many valuable contributions. This research was funded by the Social Science and Humanities Research Council of Canada and by the Social Security Administration Grant #NB17-16. This research was also supported by the U.S. Social Security Administration through grant #5-RRC08098400-10 to the National Bureau of Economic Research as part of the SSA Retirement Research Consortium. The findings and conclusions expressed are solely those of the author(s) and do not represent the views of SSA, any agency of the Federal Government, or the NBER. This project was also supported by the California Center for Population Research at UCLA (CCPR), which receives core support (P2C-HD041022) from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD). Finally, his material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No. DGE-1650604. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. All errors are our own.
labor market effects operate (Card, Kluve, and Weber 2018, Barnow and Smith 2015, Crepon and van den Berg 2016).
We re-evaluate the short- and long-run effects of means-tested employment and training programs targeted at young adults by studying the impact of the Civilian Conservation Corps (CCC). The CCC was the first and largest employment program in U.S. history and was implemented during a period of profound levels of youth unemployment—the Great Depression. Unemployment rates among young adults during the Depression were estimated to be as high as 60 percent, depending on how partial employment is counted.2 To address
high youth unemployment, the CCC was created in 1933 by the Roosevelt Administration. It employed young men aged 17 to 23 in unskilled, manual labor. Under the Army’s supervision, enrollees were sent to work in camps in rural areas where they were also fed, housed and given access to medical treatment. In addition to work experience, the CCC provided academic and vocational courses as well as cash transfers to the families of poor unemployed youths. The CCC also helped enrollees obtain employment upon completion. Enrollment in the CCC was voluntary and enlistment periods lasted 6 months with an option to re-enlist up to three times. Between 1933 and 1942, the CCC had three million enrollees and operated about 2,600 camps. Several programs in existence today such as Job Corps, Youth Conservation Corps, JobsFirstNYC, and CalWORKs are modeled after the CCC.3
We collect a new, large individual-level data set of CCC participants and their long- term outcomes. We digitize administrative records from the CCC program in Colorado and New Mexico covering the population of men in the CCC program between 1938 and 1943. Our data include dismissal records on more than 25,000 men and details their demographic characteristics, compensation, enlistment duration and reasons for leaving the program. We matched these enrollee records to 1940 Census records, WWII enlistment records, Social
2Salmond (1967) reports that in 1932, 25 percent of youths were unemployed, and another 29 percent
were only employed part-time. Rawick (1957) estimates that about 20% of youths were unemployed and another 30% were working part-time.
Security Administration records, and individual death certificates. These data allow us to investigate the effects of the CCC on important long-run outcomes and mediators including education, health, geographic mobility, employment, earnings and longevity.
To estimate the effect of the program, we exploit variation in the service duration of the enrollees. Treatment duration varied from a few days to more than two years with the average enrollee participating for approximately nine months. We show that the determinants of duration are complex and that those who trained for long periods were not necessarily from higher or lower SES backgrounds. Moreover, many ended their training for arbitrary reasons. We confirm these observations by investigating the reasons for dismissal. To assess the validity of our approach, we use the rich data from Colorado to perform some placebo tests. We find that duration does not predict pre-CCC labor outcomes or health, though we do find some effects on education. We then explicitly control for many individual and aggregate characteristics that predict participation and long-term outcomes and assess the sensitivity of our results to adding these covariates, informally and formally, as suggested by Oster (2017).
We find that individuals who trained longer in the CCC also lived longer. These gains appear to be driven by the improved health of the participants (measured by height and weight) as well as their increased geographic mobility towards richer areas, and their larger lifetime incomes. These effects are larger among Hispanics, and for those serving in times of high unemployment. We also find modest increases on educational attainment and in the probability of serving in WWII. In the short run, we find no evidence that their labor force participation, employment, or wages increased—these effects are very small and statistically insignificant. Overall, the results are consistent with the hypothesis that the program pro- vided important in-kind goods and services to disadvantaged populations in a time of need, improving their long-term health and survival. They are also consistent with the program having returns in the labor market.
of publicly available experimental data from the Job Corps (JC) program, the largest job training program in the US targeting youth with an annual budget of $1.7 billion. The JC experiment followed randomly assigned participants for four years.4 With these data, we are able to follow Lalonde (1986), using experimental data to shed light on the internal validity of a study based on observational data. Although the JC data pertains to youth training that took place in the 1990s, the program was modeled after the CCC and so retained many similar features. We focus on men that participated in the RCT for comparability. We document that JC participants are quite similar to CCC participants with regard to socio- economic characteristics (with some notable exceptions), and that they train for similar durations and quit for similar reasons.
The estimated treatment effects of training from the JC RCTs are similar in both di- rection and magnitude to the effects of duration in a simple OLS model that controls for basic observables at baseline, suggesting that our estimation strategy is internally valid. The results also speak to external validity. The original JC RCT reported that the program increases education levels, has small effects on employment rates and has positive, but statis- tically insignificant, effects on wages among those employed. We replicate these findings for men. We also document that JC and CCC both increased geographic mobility and improved health. Our results from CCC are similar in the short-term to the effects of JC, except for employment and wages.
This suggests that our long-run estimates of job training based on the CCC are likely informative about the long-run effects of JC particularly for health. There does exist a single study examining the effects of JC on labor market outcomes over 20 years using administra- tive tax data. Schochet (2018) finds no employment or earnings effects in the overall sample, though there are some positive effects for individuals who were older at baseline. They also report a 40% reduction in SSDI benefits, suggesting JC improved health, consistent with our longevity results. Using data from the Social Security Administration, we find CCC resulted
in a 3.9% increase in pension amounts, which are a function of individuals’ highest 35 years of earnings. This corresponds to an increase of roughly 6% in lifetime earnings. These effects are larger than the 2% (imprecise) increase Schochet (2018) documented, suggesting that the 20-year evaluation underestimates the returns of the program, or alternatively, that the economic conditions prevailing in the 20 years after the training took place have large effects on its return.
Our results suggest that JC participants today may live longer as a result of the program. As such, job training evaluations that focus only on the labor market impact of the program may underestimate the overall benefits. Our findings also suggest that there are in fact positive returns to investing in young adults, contrary to the commonly stated findings that returns on human capital investment are low after age 18. Our conclusion differs from that of Hendren and Sprung-Keyser (2020), who report low values for JC, because we are able to incorporate large increases in longevity, as well as increases in lifetime earnings into the benefits of the program.
This paper also contributes to the broader evaluation of the New Deal programs developed during the Great Depression. The Great Recession of 2008 renewed interest in understanding whether and for whom government programs deployed during large economic crises can be effective. Fishback (2017) provides a comprehensive survey of the literature on the effects of New Deal programs, and reports that studies show New Deal programs increased internal migration, lowered crime and reduced mortality in the short run. (See also Fishback, Haines and Kantor, 2007 and Vellore 2014.) Our results are consistent with these findings for migration and health. To our knowledge, there have not been any statistical studies of the long-term causal effects of the CCC program or of any other New Deal program on individual lifetime outcomes. Our results suggest that cost benefit analysis that do not include such outcomes may generate incorrect estimates.